Merge remote-tracking branch 'remotes/riku/tags/pull-linux-user-20141111' into staging
[qemu.git] / xen-hvm.c
1 /*
2 * Copyright (C) 2010 Citrix Ltd.
3 *
4 * This work is licensed under the terms of the GNU GPL, version 2. See
5 * the COPYING file in the top-level directory.
6 *
7 * Contributions after 2012-01-13 are licensed under the terms of the
8 * GNU GPL, version 2 or (at your option) any later version.
9 */
10
11 #include <sys/mman.h>
12
13 #include "hw/pci/pci.h"
14 #include "hw/i386/pc.h"
15 #include "hw/xen/xen_common.h"
16 #include "hw/xen/xen_backend.h"
17 #include "qmp-commands.h"
18
19 #include "sysemu/char.h"
20 #include "qemu/range.h"
21 #include "sysemu/xen-mapcache.h"
22 #include "trace.h"
23 #include "exec/address-spaces.h"
24
25 #include <xen/hvm/ioreq.h>
26 #include <xen/hvm/params.h>
27 #include <xen/hvm/e820.h>
28
29 //#define DEBUG_XEN_HVM
30
31 #ifdef DEBUG_XEN_HVM
32 #define DPRINTF(fmt, ...) \
33 do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
34 #else
35 #define DPRINTF(fmt, ...) \
36 do { } while (0)
37 #endif
38
39 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
40 static MemoryRegion *framebuffer;
41 static bool xen_in_migration;
42
43 /* Compatibility with older version */
44
45 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
46 * installed. This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
47 * needs to be included before this block and hw/xen/xen_common.h needs to
48 * be included before xen/hvm/ioreq.h
49 */
50 #ifndef IOREQ_TYPE_VMWARE_PORT
51 #define IOREQ_TYPE_VMWARE_PORT 3
52 struct vmware_regs {
53 uint32_t esi;
54 uint32_t edi;
55 uint32_t ebx;
56 uint32_t ecx;
57 uint32_t edx;
58 };
59 typedef struct vmware_regs vmware_regs_t;
60
61 struct shared_vmport_iopage {
62 struct vmware_regs vcpu_vmport_regs[1];
63 };
64 typedef struct shared_vmport_iopage shared_vmport_iopage_t;
65 #endif
66
67 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
68 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
69 {
70 return shared_page->vcpu_iodata[i].vp_eport;
71 }
72 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
73 {
74 return &shared_page->vcpu_iodata[vcpu].vp_ioreq;
75 }
76 # define FMT_ioreq_size PRIx64
77 #else
78 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
79 {
80 return shared_page->vcpu_ioreq[i].vp_eport;
81 }
82 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
83 {
84 return &shared_page->vcpu_ioreq[vcpu];
85 }
86 # define FMT_ioreq_size "u"
87 #endif
88 #ifndef HVM_PARAM_BUFIOREQ_EVTCHN
89 #define HVM_PARAM_BUFIOREQ_EVTCHN 26
90 #endif
91
92 #define BUFFER_IO_MAX_DELAY 100
93
94 typedef struct XenPhysmap {
95 hwaddr start_addr;
96 ram_addr_t size;
97 const char *name;
98 hwaddr phys_offset;
99
100 QLIST_ENTRY(XenPhysmap) list;
101 } XenPhysmap;
102
103 typedef struct XenIOState {
104 shared_iopage_t *shared_page;
105 shared_vmport_iopage_t *shared_vmport_page;
106 buffered_iopage_t *buffered_io_page;
107 QEMUTimer *buffered_io_timer;
108 CPUState **cpu_by_vcpu_id;
109 /* the evtchn port for polling the notification, */
110 evtchn_port_t *ioreq_local_port;
111 /* evtchn local port for buffered io */
112 evtchn_port_t bufioreq_local_port;
113 /* the evtchn fd for polling */
114 XenEvtchn xce_handle;
115 /* which vcpu we are serving */
116 int send_vcpu;
117
118 struct xs_handle *xenstore;
119 MemoryListener memory_listener;
120 QLIST_HEAD(, XenPhysmap) physmap;
121 hwaddr free_phys_offset;
122 const XenPhysmap *log_for_dirtybit;
123
124 Notifier exit;
125 Notifier suspend;
126 Notifier wakeup;
127 } XenIOState;
128
129 /* Xen specific function for piix pci */
130
131 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
132 {
133 return irq_num + ((pci_dev->devfn >> 3) << 2);
134 }
135
136 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
137 {
138 xc_hvm_set_pci_intx_level(xen_xc, xen_domid, 0, 0, irq_num >> 2,
139 irq_num & 3, level);
140 }
141
142 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
143 {
144 int i;
145
146 /* Scan for updates to PCI link routes (0x60-0x63). */
147 for (i = 0; i < len; i++) {
148 uint8_t v = (val >> (8 * i)) & 0xff;
149 if (v & 0x80) {
150 v = 0;
151 }
152 v &= 0xf;
153 if (((address + i) >= 0x60) && ((address + i) <= 0x63)) {
154 xc_hvm_set_pci_link_route(xen_xc, xen_domid, address + i - 0x60, v);
155 }
156 }
157 }
158
159 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
160 {
161 xen_xc_hvm_inject_msi(xen_xc, xen_domid, addr, data);
162 }
163
164 static void xen_suspend_notifier(Notifier *notifier, void *data)
165 {
166 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
167 }
168
169 /* Xen Interrupt Controller */
170
171 static void xen_set_irq(void *opaque, int irq, int level)
172 {
173 xc_hvm_set_isa_irq_level(xen_xc, xen_domid, irq, level);
174 }
175
176 qemu_irq *xen_interrupt_controller_init(void)
177 {
178 return qemu_allocate_irqs(xen_set_irq, NULL, 16);
179 }
180
181 /* Memory Ops */
182
183 static void xen_ram_init(ram_addr_t *below_4g_mem_size,
184 ram_addr_t *above_4g_mem_size,
185 ram_addr_t ram_size, MemoryRegion **ram_memory_p)
186 {
187 MemoryRegion *sysmem = get_system_memory();
188 ram_addr_t block_len;
189 uint64_t user_lowmem = object_property_get_int(qdev_get_machine(),
190 PC_MACHINE_MAX_RAM_BELOW_4G,
191 &error_abort);
192
193 /* Handle the machine opt max-ram-below-4g. It is basically doing
194 * min(xen limit, user limit).
195 */
196 if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
197 user_lowmem = HVM_BELOW_4G_RAM_END;
198 }
199
200 if (ram_size >= user_lowmem) {
201 *above_4g_mem_size = ram_size - user_lowmem;
202 *below_4g_mem_size = user_lowmem;
203 } else {
204 *above_4g_mem_size = 0;
205 *below_4g_mem_size = ram_size;
206 }
207 if (!*above_4g_mem_size) {
208 block_len = ram_size;
209 } else {
210 /*
211 * Xen does not allocate the memory continuously, it keeps a
212 * hole of the size computed above or passed in.
213 */
214 block_len = (1ULL << 32) + *above_4g_mem_size;
215 }
216 memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
217 &error_abort);
218 *ram_memory_p = &ram_memory;
219 vmstate_register_ram_global(&ram_memory);
220
221 memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
222 &ram_memory, 0, 0xa0000);
223 memory_region_add_subregion(sysmem, 0, &ram_640k);
224 /* Skip of the VGA IO memory space, it will be registered later by the VGA
225 * emulated device.
226 *
227 * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
228 * the Options ROM, so it is registered here as RAM.
229 */
230 memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
231 &ram_memory, 0xc0000,
232 *below_4g_mem_size - 0xc0000);
233 memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
234 if (*above_4g_mem_size > 0) {
235 memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
236 &ram_memory, 0x100000000ULL,
237 *above_4g_mem_size);
238 memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
239 }
240 }
241
242 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
243 {
244 unsigned long nr_pfn;
245 xen_pfn_t *pfn_list;
246 int i;
247
248 if (runstate_check(RUN_STATE_INMIGRATE)) {
249 /* RAM already populated in Xen */
250 fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
251 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
252 __func__, size, ram_addr);
253 return;
254 }
255
256 if (mr == &ram_memory) {
257 return;
258 }
259
260 trace_xen_ram_alloc(ram_addr, size);
261
262 nr_pfn = size >> TARGET_PAGE_BITS;
263 pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
264
265 for (i = 0; i < nr_pfn; i++) {
266 pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
267 }
268
269 if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
270 hw_error("xen: failed to populate ram at " RAM_ADDR_FMT, ram_addr);
271 }
272
273 g_free(pfn_list);
274 }
275
276 static XenPhysmap *get_physmapping(XenIOState *state,
277 hwaddr start_addr, ram_addr_t size)
278 {
279 XenPhysmap *physmap = NULL;
280
281 start_addr &= TARGET_PAGE_MASK;
282
283 QLIST_FOREACH(physmap, &state->physmap, list) {
284 if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
285 return physmap;
286 }
287 }
288 return NULL;
289 }
290
291 static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
292 ram_addr_t size, void *opaque)
293 {
294 hwaddr addr = start_addr & TARGET_PAGE_MASK;
295 XenIOState *xen_io_state = opaque;
296 XenPhysmap *physmap = NULL;
297
298 QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
299 if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
300 return physmap->start_addr;
301 }
302 }
303
304 return start_addr;
305 }
306
307 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
308 static int xen_add_to_physmap(XenIOState *state,
309 hwaddr start_addr,
310 ram_addr_t size,
311 MemoryRegion *mr,
312 hwaddr offset_within_region)
313 {
314 unsigned long i = 0;
315 int rc = 0;
316 XenPhysmap *physmap = NULL;
317 hwaddr pfn, start_gpfn;
318 hwaddr phys_offset = memory_region_get_ram_addr(mr);
319 char path[80], value[17];
320 const char *mr_name;
321
322 if (get_physmapping(state, start_addr, size)) {
323 return 0;
324 }
325 if (size <= 0) {
326 return -1;
327 }
328
329 /* Xen can only handle a single dirty log region for now and we want
330 * the linear framebuffer to be that region.
331 * Avoid tracking any regions that is not videoram and avoid tracking
332 * the legacy vga region. */
333 if (mr == framebuffer && start_addr > 0xbffff) {
334 goto go_physmap;
335 }
336 return -1;
337
338 go_physmap:
339 DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
340 start_addr, start_addr + size);
341
342 pfn = phys_offset >> TARGET_PAGE_BITS;
343 start_gpfn = start_addr >> TARGET_PAGE_BITS;
344 for (i = 0; i < size >> TARGET_PAGE_BITS; i++) {
345 unsigned long idx = pfn + i;
346 xen_pfn_t gpfn = start_gpfn + i;
347
348 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
349 if (rc) {
350 DPRINTF("add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
351 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
352 return -rc;
353 }
354 }
355
356 mr_name = memory_region_name(mr);
357
358 physmap = g_malloc(sizeof (XenPhysmap));
359
360 physmap->start_addr = start_addr;
361 physmap->size = size;
362 physmap->name = mr_name;
363 physmap->phys_offset = phys_offset;
364
365 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
366
367 xc_domain_pin_memory_cacheattr(xen_xc, xen_domid,
368 start_addr >> TARGET_PAGE_BITS,
369 (start_addr + size - 1) >> TARGET_PAGE_BITS,
370 XEN_DOMCTL_MEM_CACHEATTR_WB);
371
372 snprintf(path, sizeof(path),
373 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
374 xen_domid, (uint64_t)phys_offset);
375 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)start_addr);
376 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
377 return -1;
378 }
379 snprintf(path, sizeof(path),
380 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
381 xen_domid, (uint64_t)phys_offset);
382 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)size);
383 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
384 return -1;
385 }
386 if (mr_name) {
387 snprintf(path, sizeof(path),
388 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
389 xen_domid, (uint64_t)phys_offset);
390 if (!xs_write(state->xenstore, 0, path, mr_name, strlen(mr_name))) {
391 return -1;
392 }
393 }
394
395 return 0;
396 }
397
398 static int xen_remove_from_physmap(XenIOState *state,
399 hwaddr start_addr,
400 ram_addr_t size)
401 {
402 unsigned long i = 0;
403 int rc = 0;
404 XenPhysmap *physmap = NULL;
405 hwaddr phys_offset = 0;
406
407 physmap = get_physmapping(state, start_addr, size);
408 if (physmap == NULL) {
409 return -1;
410 }
411
412 phys_offset = physmap->phys_offset;
413 size = physmap->size;
414
415 DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
416 "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
417
418 size >>= TARGET_PAGE_BITS;
419 start_addr >>= TARGET_PAGE_BITS;
420 phys_offset >>= TARGET_PAGE_BITS;
421 for (i = 0; i < size; i++) {
422 xen_pfn_t idx = start_addr + i;
423 xen_pfn_t gpfn = phys_offset + i;
424
425 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
426 if (rc) {
427 fprintf(stderr, "add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
428 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
429 return -rc;
430 }
431 }
432
433 QLIST_REMOVE(physmap, list);
434 if (state->log_for_dirtybit == physmap) {
435 state->log_for_dirtybit = NULL;
436 }
437 g_free(physmap);
438
439 return 0;
440 }
441
442 #else
443 static int xen_add_to_physmap(XenIOState *state,
444 hwaddr start_addr,
445 ram_addr_t size,
446 MemoryRegion *mr,
447 hwaddr offset_within_region)
448 {
449 return -ENOSYS;
450 }
451
452 static int xen_remove_from_physmap(XenIOState *state,
453 hwaddr start_addr,
454 ram_addr_t size)
455 {
456 return -ENOSYS;
457 }
458 #endif
459
460 static void xen_set_memory(struct MemoryListener *listener,
461 MemoryRegionSection *section,
462 bool add)
463 {
464 XenIOState *state = container_of(listener, XenIOState, memory_listener);
465 hwaddr start_addr = section->offset_within_address_space;
466 ram_addr_t size = int128_get64(section->size);
467 bool log_dirty = memory_region_is_logging(section->mr);
468 hvmmem_type_t mem_type;
469
470 if (!memory_region_is_ram(section->mr)) {
471 return;
472 }
473
474 if (!(section->mr != &ram_memory
475 && ( (log_dirty && add) || (!log_dirty && !add)))) {
476 return;
477 }
478
479 trace_xen_client_set_memory(start_addr, size, log_dirty);
480
481 start_addr &= TARGET_PAGE_MASK;
482 size = TARGET_PAGE_ALIGN(size);
483
484 if (add) {
485 if (!memory_region_is_rom(section->mr)) {
486 xen_add_to_physmap(state, start_addr, size,
487 section->mr, section->offset_within_region);
488 } else {
489 mem_type = HVMMEM_ram_ro;
490 if (xc_hvm_set_mem_type(xen_xc, xen_domid, mem_type,
491 start_addr >> TARGET_PAGE_BITS,
492 size >> TARGET_PAGE_BITS)) {
493 DPRINTF("xc_hvm_set_mem_type error, addr: "TARGET_FMT_plx"\n",
494 start_addr);
495 }
496 }
497 } else {
498 if (xen_remove_from_physmap(state, start_addr, size) < 0) {
499 DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
500 }
501 }
502 }
503
504 static void xen_region_add(MemoryListener *listener,
505 MemoryRegionSection *section)
506 {
507 memory_region_ref(section->mr);
508 xen_set_memory(listener, section, true);
509 }
510
511 static void xen_region_del(MemoryListener *listener,
512 MemoryRegionSection *section)
513 {
514 xen_set_memory(listener, section, false);
515 memory_region_unref(section->mr);
516 }
517
518 static void xen_sync_dirty_bitmap(XenIOState *state,
519 hwaddr start_addr,
520 ram_addr_t size)
521 {
522 hwaddr npages = size >> TARGET_PAGE_BITS;
523 const int width = sizeof(unsigned long) * 8;
524 unsigned long bitmap[(npages + width - 1) / width];
525 int rc, i, j;
526 const XenPhysmap *physmap = NULL;
527
528 physmap = get_physmapping(state, start_addr, size);
529 if (physmap == NULL) {
530 /* not handled */
531 return;
532 }
533
534 if (state->log_for_dirtybit == NULL) {
535 state->log_for_dirtybit = physmap;
536 } else if (state->log_for_dirtybit != physmap) {
537 /* Only one range for dirty bitmap can be tracked. */
538 return;
539 }
540
541 rc = xc_hvm_track_dirty_vram(xen_xc, xen_domid,
542 start_addr >> TARGET_PAGE_BITS, npages,
543 bitmap);
544 if (rc < 0) {
545 #ifndef ENODATA
546 #define ENODATA ENOENT
547 #endif
548 if (errno == ENODATA) {
549 memory_region_set_dirty(framebuffer, 0, size);
550 DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
551 ", 0x" TARGET_FMT_plx "): %s\n",
552 start_addr, start_addr + size, strerror(errno));
553 }
554 return;
555 }
556
557 for (i = 0; i < ARRAY_SIZE(bitmap); i++) {
558 unsigned long map = bitmap[i];
559 while (map != 0) {
560 j = ctzl(map);
561 map &= ~(1ul << j);
562 memory_region_set_dirty(framebuffer,
563 (i * width + j) * TARGET_PAGE_SIZE,
564 TARGET_PAGE_SIZE);
565 };
566 }
567 }
568
569 static void xen_log_start(MemoryListener *listener,
570 MemoryRegionSection *section)
571 {
572 XenIOState *state = container_of(listener, XenIOState, memory_listener);
573
574 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
575 int128_get64(section->size));
576 }
577
578 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section)
579 {
580 XenIOState *state = container_of(listener, XenIOState, memory_listener);
581
582 state->log_for_dirtybit = NULL;
583 /* Disable dirty bit tracking */
584 xc_hvm_track_dirty_vram(xen_xc, xen_domid, 0, 0, NULL);
585 }
586
587 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
588 {
589 XenIOState *state = container_of(listener, XenIOState, memory_listener);
590
591 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
592 int128_get64(section->size));
593 }
594
595 static void xen_log_global_start(MemoryListener *listener)
596 {
597 if (xen_enabled()) {
598 xen_in_migration = true;
599 }
600 }
601
602 static void xen_log_global_stop(MemoryListener *listener)
603 {
604 xen_in_migration = false;
605 }
606
607 static MemoryListener xen_memory_listener = {
608 .region_add = xen_region_add,
609 .region_del = xen_region_del,
610 .log_start = xen_log_start,
611 .log_stop = xen_log_stop,
612 .log_sync = xen_log_sync,
613 .log_global_start = xen_log_global_start,
614 .log_global_stop = xen_log_global_stop,
615 .priority = 10,
616 };
617
618 /* get the ioreq packets from share mem */
619 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
620 {
621 ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
622
623 if (req->state != STATE_IOREQ_READY) {
624 DPRINTF("I/O request not ready: "
625 "%x, ptr: %x, port: %"PRIx64", "
626 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
627 req->state, req->data_is_ptr, req->addr,
628 req->data, req->count, req->size);
629 return NULL;
630 }
631
632 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
633
634 req->state = STATE_IOREQ_INPROCESS;
635 return req;
636 }
637
638 /* use poll to get the port notification */
639 /* ioreq_vec--out,the */
640 /* retval--the number of ioreq packet */
641 static ioreq_t *cpu_get_ioreq(XenIOState *state)
642 {
643 int i;
644 evtchn_port_t port;
645
646 port = xc_evtchn_pending(state->xce_handle);
647 if (port == state->bufioreq_local_port) {
648 timer_mod(state->buffered_io_timer,
649 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
650 return NULL;
651 }
652
653 if (port != -1) {
654 for (i = 0; i < max_cpus; i++) {
655 if (state->ioreq_local_port[i] == port) {
656 break;
657 }
658 }
659
660 if (i == max_cpus) {
661 hw_error("Fatal error while trying to get io event!\n");
662 }
663
664 /* unmask the wanted port again */
665 xc_evtchn_unmask(state->xce_handle, port);
666
667 /* get the io packet from shared memory */
668 state->send_vcpu = i;
669 return cpu_get_ioreq_from_shared_memory(state, i);
670 }
671
672 /* read error or read nothing */
673 return NULL;
674 }
675
676 static uint32_t do_inp(pio_addr_t addr, unsigned long size)
677 {
678 switch (size) {
679 case 1:
680 return cpu_inb(addr);
681 case 2:
682 return cpu_inw(addr);
683 case 4:
684 return cpu_inl(addr);
685 default:
686 hw_error("inp: bad size: %04"FMT_pioaddr" %lx", addr, size);
687 }
688 }
689
690 static void do_outp(pio_addr_t addr,
691 unsigned long size, uint32_t val)
692 {
693 switch (size) {
694 case 1:
695 return cpu_outb(addr, val);
696 case 2:
697 return cpu_outw(addr, val);
698 case 4:
699 return cpu_outl(addr, val);
700 default:
701 hw_error("outp: bad size: %04"FMT_pioaddr" %lx", addr, size);
702 }
703 }
704
705 /*
706 * Helper functions which read/write an object from/to physical guest
707 * memory, as part of the implementation of an ioreq.
708 *
709 * Equivalent to
710 * cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
711 * val, req->size, 0/1)
712 * except without the integer overflow problems.
713 */
714 static void rw_phys_req_item(hwaddr addr,
715 ioreq_t *req, uint32_t i, void *val, int rw)
716 {
717 /* Do everything unsigned so overflow just results in a truncated result
718 * and accesses to undesired parts of guest memory, which is up
719 * to the guest */
720 hwaddr offset = (hwaddr)req->size * i;
721 if (req->df) {
722 addr -= offset;
723 } else {
724 addr += offset;
725 }
726 cpu_physical_memory_rw(addr, val, req->size, rw);
727 }
728
729 static inline void read_phys_req_item(hwaddr addr,
730 ioreq_t *req, uint32_t i, void *val)
731 {
732 rw_phys_req_item(addr, req, i, val, 0);
733 }
734 static inline void write_phys_req_item(hwaddr addr,
735 ioreq_t *req, uint32_t i, void *val)
736 {
737 rw_phys_req_item(addr, req, i, val, 1);
738 }
739
740
741 static void cpu_ioreq_pio(ioreq_t *req)
742 {
743 uint32_t i;
744
745 if (req->dir == IOREQ_READ) {
746 if (!req->data_is_ptr) {
747 req->data = do_inp(req->addr, req->size);
748 } else {
749 uint32_t tmp;
750
751 for (i = 0; i < req->count; i++) {
752 tmp = do_inp(req->addr, req->size);
753 write_phys_req_item(req->data, req, i, &tmp);
754 }
755 }
756 } else if (req->dir == IOREQ_WRITE) {
757 if (!req->data_is_ptr) {
758 do_outp(req->addr, req->size, req->data);
759 } else {
760 for (i = 0; i < req->count; i++) {
761 uint32_t tmp = 0;
762
763 read_phys_req_item(req->data, req, i, &tmp);
764 do_outp(req->addr, req->size, tmp);
765 }
766 }
767 }
768 }
769
770 static void cpu_ioreq_move(ioreq_t *req)
771 {
772 uint32_t i;
773
774 if (!req->data_is_ptr) {
775 if (req->dir == IOREQ_READ) {
776 for (i = 0; i < req->count; i++) {
777 read_phys_req_item(req->addr, req, i, &req->data);
778 }
779 } else if (req->dir == IOREQ_WRITE) {
780 for (i = 0; i < req->count; i++) {
781 write_phys_req_item(req->addr, req, i, &req->data);
782 }
783 }
784 } else {
785 uint64_t tmp;
786
787 if (req->dir == IOREQ_READ) {
788 for (i = 0; i < req->count; i++) {
789 read_phys_req_item(req->addr, req, i, &tmp);
790 write_phys_req_item(req->data, req, i, &tmp);
791 }
792 } else if (req->dir == IOREQ_WRITE) {
793 for (i = 0; i < req->count; i++) {
794 read_phys_req_item(req->data, req, i, &tmp);
795 write_phys_req_item(req->addr, req, i, &tmp);
796 }
797 }
798 }
799 }
800
801 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
802 {
803 X86CPU *cpu;
804 CPUX86State *env;
805
806 cpu = X86_CPU(current_cpu);
807 env = &cpu->env;
808 env->regs[R_EAX] = req->data;
809 env->regs[R_EBX] = vmport_regs->ebx;
810 env->regs[R_ECX] = vmport_regs->ecx;
811 env->regs[R_EDX] = vmport_regs->edx;
812 env->regs[R_ESI] = vmport_regs->esi;
813 env->regs[R_EDI] = vmport_regs->edi;
814 }
815
816 static void regs_from_cpu(vmware_regs_t *vmport_regs)
817 {
818 X86CPU *cpu = X86_CPU(current_cpu);
819 CPUX86State *env = &cpu->env;
820
821 vmport_regs->ebx = env->regs[R_EBX];
822 vmport_regs->ecx = env->regs[R_ECX];
823 vmport_regs->edx = env->regs[R_EDX];
824 vmport_regs->esi = env->regs[R_ESI];
825 vmport_regs->edi = env->regs[R_EDI];
826 }
827
828 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
829 {
830 vmware_regs_t *vmport_regs;
831
832 assert(state->shared_vmport_page);
833 vmport_regs =
834 &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
835 QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
836
837 current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
838 regs_to_cpu(vmport_regs, req);
839 cpu_ioreq_pio(req);
840 regs_from_cpu(vmport_regs);
841 current_cpu = NULL;
842 }
843
844 static void handle_ioreq(XenIOState *state, ioreq_t *req)
845 {
846 if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
847 (req->size < sizeof (target_ulong))) {
848 req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
849 }
850
851 switch (req->type) {
852 case IOREQ_TYPE_PIO:
853 cpu_ioreq_pio(req);
854 break;
855 case IOREQ_TYPE_COPY:
856 cpu_ioreq_move(req);
857 break;
858 case IOREQ_TYPE_VMWARE_PORT:
859 handle_vmport_ioreq(state, req);
860 break;
861 case IOREQ_TYPE_TIMEOFFSET:
862 break;
863 case IOREQ_TYPE_INVALIDATE:
864 xen_invalidate_map_cache();
865 break;
866 default:
867 hw_error("Invalid ioreq type 0x%x\n", req->type);
868 }
869 }
870
871 static int handle_buffered_iopage(XenIOState *state)
872 {
873 buf_ioreq_t *buf_req = NULL;
874 ioreq_t req;
875 int qw;
876
877 if (!state->buffered_io_page) {
878 return 0;
879 }
880
881 memset(&req, 0x00, sizeof(req));
882
883 while (state->buffered_io_page->read_pointer != state->buffered_io_page->write_pointer) {
884 buf_req = &state->buffered_io_page->buf_ioreq[
885 state->buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
886 req.size = 1UL << buf_req->size;
887 req.count = 1;
888 req.addr = buf_req->addr;
889 req.data = buf_req->data;
890 req.state = STATE_IOREQ_READY;
891 req.dir = buf_req->dir;
892 req.df = 1;
893 req.type = buf_req->type;
894 req.data_is_ptr = 0;
895 qw = (req.size == 8);
896 if (qw) {
897 buf_req = &state->buffered_io_page->buf_ioreq[
898 (state->buffered_io_page->read_pointer + 1) % IOREQ_BUFFER_SLOT_NUM];
899 req.data |= ((uint64_t)buf_req->data) << 32;
900 }
901
902 handle_ioreq(state, &req);
903
904 xen_mb();
905 state->buffered_io_page->read_pointer += qw ? 2 : 1;
906 }
907
908 return req.count;
909 }
910
911 static void handle_buffered_io(void *opaque)
912 {
913 XenIOState *state = opaque;
914
915 if (handle_buffered_iopage(state)) {
916 timer_mod(state->buffered_io_timer,
917 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
918 } else {
919 timer_del(state->buffered_io_timer);
920 xc_evtchn_unmask(state->xce_handle, state->bufioreq_local_port);
921 }
922 }
923
924 static void cpu_handle_ioreq(void *opaque)
925 {
926 XenIOState *state = opaque;
927 ioreq_t *req = cpu_get_ioreq(state);
928
929 handle_buffered_iopage(state);
930 if (req) {
931 handle_ioreq(state, req);
932
933 if (req->state != STATE_IOREQ_INPROCESS) {
934 fprintf(stderr, "Badness in I/O request ... not in service?!: "
935 "%x, ptr: %x, port: %"PRIx64", "
936 "data: %"PRIx64", count: %" FMT_ioreq_size
937 ", size: %" FMT_ioreq_size
938 ", type: %"FMT_ioreq_size"\n",
939 req->state, req->data_is_ptr, req->addr,
940 req->data, req->count, req->size, req->type);
941 destroy_hvm_domain(false);
942 return;
943 }
944
945 xen_wmb(); /* Update ioreq contents /then/ update state. */
946
947 /*
948 * We do this before we send the response so that the tools
949 * have the opportunity to pick up on the reset before the
950 * guest resumes and does a hlt with interrupts disabled which
951 * causes Xen to powerdown the domain.
952 */
953 if (runstate_is_running()) {
954 if (qemu_shutdown_requested_get()) {
955 destroy_hvm_domain(false);
956 }
957 if (qemu_reset_requested_get()) {
958 qemu_system_reset(VMRESET_REPORT);
959 destroy_hvm_domain(true);
960 }
961 }
962
963 req->state = STATE_IORESP_READY;
964 xc_evtchn_notify(state->xce_handle, state->ioreq_local_port[state->send_vcpu]);
965 }
966 }
967
968 static void xen_main_loop_prepare(XenIOState *state)
969 {
970 int evtchn_fd = -1;
971
972 if (state->xce_handle != XC_HANDLER_INITIAL_VALUE) {
973 evtchn_fd = xc_evtchn_fd(state->xce_handle);
974 }
975
976 state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
977 state);
978
979 if (evtchn_fd != -1) {
980 CPUState *cpu_state;
981
982 DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
983 CPU_FOREACH(cpu_state) {
984 DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
985 __func__, cpu_state->cpu_index, cpu_state);
986 state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
987 }
988 qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
989 }
990 }
991
992
993 static void xen_hvm_change_state_handler(void *opaque, int running,
994 RunState rstate)
995 {
996 XenIOState *xstate = opaque;
997 if (running) {
998 xen_main_loop_prepare(xstate);
999 }
1000 }
1001
1002 static void xen_exit_notifier(Notifier *n, void *data)
1003 {
1004 XenIOState *state = container_of(n, XenIOState, exit);
1005
1006 xc_evtchn_close(state->xce_handle);
1007 xs_daemon_close(state->xenstore);
1008 }
1009
1010 static void xen_read_physmap(XenIOState *state)
1011 {
1012 XenPhysmap *physmap = NULL;
1013 unsigned int len, num, i;
1014 char path[80], *value = NULL;
1015 char **entries = NULL;
1016
1017 snprintf(path, sizeof(path),
1018 "/local/domain/0/device-model/%d/physmap", xen_domid);
1019 entries = xs_directory(state->xenstore, 0, path, &num);
1020 if (entries == NULL)
1021 return;
1022
1023 for (i = 0; i < num; i++) {
1024 physmap = g_malloc(sizeof (XenPhysmap));
1025 physmap->phys_offset = strtoull(entries[i], NULL, 16);
1026 snprintf(path, sizeof(path),
1027 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1028 xen_domid, entries[i]);
1029 value = xs_read(state->xenstore, 0, path, &len);
1030 if (value == NULL) {
1031 g_free(physmap);
1032 continue;
1033 }
1034 physmap->start_addr = strtoull(value, NULL, 16);
1035 free(value);
1036
1037 snprintf(path, sizeof(path),
1038 "/local/domain/0/device-model/%d/physmap/%s/size",
1039 xen_domid, entries[i]);
1040 value = xs_read(state->xenstore, 0, path, &len);
1041 if (value == NULL) {
1042 g_free(physmap);
1043 continue;
1044 }
1045 physmap->size = strtoull(value, NULL, 16);
1046 free(value);
1047
1048 snprintf(path, sizeof(path),
1049 "/local/domain/0/device-model/%d/physmap/%s/name",
1050 xen_domid, entries[i]);
1051 physmap->name = xs_read(state->xenstore, 0, path, &len);
1052
1053 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
1054 }
1055 free(entries);
1056 }
1057
1058 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1059 {
1060 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1061 }
1062
1063 /* return 0 means OK, or -1 means critical issue -- will exit(1) */
1064 int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
1065 MemoryRegion **ram_memory)
1066 {
1067 int i, rc;
1068 unsigned long ioreq_pfn;
1069 unsigned long bufioreq_evtchn;
1070 XenIOState *state;
1071
1072 state = g_malloc0(sizeof (XenIOState));
1073
1074 state->xce_handle = xen_xc_evtchn_open(NULL, 0);
1075 if (state->xce_handle == XC_HANDLER_INITIAL_VALUE) {
1076 perror("xen: event channel open");
1077 return -1;
1078 }
1079
1080 state->xenstore = xs_daemon_open();
1081 if (state->xenstore == NULL) {
1082 perror("xen: xenstore open");
1083 return -1;
1084 }
1085
1086 state->exit.notify = xen_exit_notifier;
1087 qemu_add_exit_notifier(&state->exit);
1088
1089 state->suspend.notify = xen_suspend_notifier;
1090 qemu_register_suspend_notifier(&state->suspend);
1091
1092 state->wakeup.notify = xen_wakeup_notifier;
1093 qemu_register_wakeup_notifier(&state->wakeup);
1094
1095 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_IOREQ_PFN, &ioreq_pfn);
1096 DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1097 state->shared_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1098 PROT_READ|PROT_WRITE, ioreq_pfn);
1099 if (state->shared_page == NULL) {
1100 hw_error("map shared IO page returned error %d handle=" XC_INTERFACE_FMT,
1101 errno, xen_xc);
1102 }
1103
1104 rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1105 if (!rc) {
1106 DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1107 state->shared_vmport_page =
1108 xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1109 PROT_READ|PROT_WRITE, ioreq_pfn);
1110 if (state->shared_vmport_page == NULL) {
1111 hw_error("map shared vmport IO page returned error %d handle="
1112 XC_INTERFACE_FMT, errno, xen_xc);
1113 }
1114 } else if (rc != -ENOSYS) {
1115 hw_error("get vmport regs pfn returned error %d, rc=%d", errno, rc);
1116 }
1117
1118 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
1119 DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
1120 state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1121 PROT_READ|PROT_WRITE, ioreq_pfn);
1122 if (state->buffered_io_page == NULL) {
1123 hw_error("map buffered IO page returned error %d", errno);
1124 }
1125
1126 /* Note: cpus is empty at this point in init */
1127 state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1128
1129 state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1130
1131 /* FIXME: how about if we overflow the page here? */
1132 for (i = 0; i < max_cpus; i++) {
1133 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1134 xen_vcpu_eport(state->shared_page, i));
1135 if (rc == -1) {
1136 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1137 return -1;
1138 }
1139 state->ioreq_local_port[i] = rc;
1140 }
1141
1142 rc = xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_EVTCHN,
1143 &bufioreq_evtchn);
1144 if (rc < 0) {
1145 fprintf(stderr, "failed to get HVM_PARAM_BUFIOREQ_EVTCHN\n");
1146 return -1;
1147 }
1148 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1149 (uint32_t)bufioreq_evtchn);
1150 if (rc == -1) {
1151 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1152 return -1;
1153 }
1154 state->bufioreq_local_port = rc;
1155
1156 /* Init RAM management */
1157 xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1158 xen_ram_init(below_4g_mem_size, above_4g_mem_size, ram_size, ram_memory);
1159
1160 qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1161
1162 state->memory_listener = xen_memory_listener;
1163 QLIST_INIT(&state->physmap);
1164 memory_listener_register(&state->memory_listener, &address_space_memory);
1165 state->log_for_dirtybit = NULL;
1166
1167 /* Initialize backend core & drivers */
1168 if (xen_be_init() != 0) {
1169 fprintf(stderr, "%s: xen backend core setup failed\n", __FUNCTION__);
1170 return -1;
1171 }
1172 xen_be_register("console", &xen_console_ops);
1173 xen_be_register("vkbd", &xen_kbdmouse_ops);
1174 xen_be_register("qdisk", &xen_blkdev_ops);
1175 xen_read_physmap(state);
1176
1177 return 0;
1178 }
1179
1180 void destroy_hvm_domain(bool reboot)
1181 {
1182 XenXC xc_handle;
1183 int sts;
1184
1185 xc_handle = xen_xc_interface_open(0, 0, 0);
1186 if (xc_handle == XC_HANDLER_INITIAL_VALUE) {
1187 fprintf(stderr, "Cannot acquire xenctrl handle\n");
1188 } else {
1189 sts = xc_domain_shutdown(xc_handle, xen_domid,
1190 reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
1191 if (sts != 0) {
1192 fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1193 "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1194 sts, strerror(errno));
1195 } else {
1196 fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1197 reboot ? "reboot" : "poweroff");
1198 }
1199 xc_interface_close(xc_handle);
1200 }
1201 }
1202
1203 void xen_register_framebuffer(MemoryRegion *mr)
1204 {
1205 framebuffer = mr;
1206 }
1207
1208 void xen_shutdown_fatal_error(const char *fmt, ...)
1209 {
1210 va_list ap;
1211
1212 va_start(ap, fmt);
1213 vfprintf(stderr, fmt, ap);
1214 va_end(ap);
1215 fprintf(stderr, "Will destroy the domain.\n");
1216 /* destroy the domain */
1217 qemu_system_shutdown_request();
1218 }
1219
1220 void xen_modified_memory(ram_addr_t start, ram_addr_t length)
1221 {
1222 if (unlikely(xen_in_migration)) {
1223 int rc;
1224 ram_addr_t start_pfn, nb_pages;
1225
1226 if (length == 0) {
1227 length = TARGET_PAGE_SIZE;
1228 }
1229 start_pfn = start >> TARGET_PAGE_BITS;
1230 nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1231 - start_pfn;
1232 rc = xc_hvm_modified_memory(xen_xc, xen_domid, start_pfn, nb_pages);
1233 if (rc) {
1234 fprintf(stderr,
1235 "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1236 __func__, start, nb_pages, rc, strerror(-rc));
1237 }
1238 }
1239 }
1240
1241 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1242 {
1243 if (enable) {
1244 memory_global_dirty_log_start();
1245 } else {
1246 memory_global_dirty_log_stop();
1247 }
1248 }